We ensure that the sampling function is explicitly called during the image generation process after obtaining z_mean and z_log_var.

Generate different images each time by explicitly calling the sampling function with new random noise.

import numpy as np
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers
import matplotlib.pyplot as plt

The process of sampling the latent vector involves generating a random sample from the distribution defined by 𝑧 mean and z log_var. This is done using the reparameterization trick, which allows gradients to flow through the sampling process during training.

# Sampling function with randomness
def sampling(z_mean, z_log_var):
    batch = tf.shape(z_mean)[0]
    dim = tf.shape(z_mean)[1]
    epsilon = tf.keras.backend.random_normal(shape=(batch, dim))
    return z_mean + tf.exp(0.5 * z_log_var) * epsilon

# VAE Class with custom call method
class VAE(keras.Model):
    def __init__(self, encoder, decoder, **kwargs):
        super(VAE, self).__init__(**kwargs)
        self.encoder = encoder
        self.decoder = decoder

    def call(self, inputs):
        z_mean, z_log_var, z = self.encoder(inputs)
        reconstructed = self.decoder(z)
        reconstruction_loss = tf.reduce_mean(
            tf.keras.losses.binary_crossentropy(inputs, reconstructed)
        )
        reconstruction_loss *= 200 * 200 * 3
        kl_loss = 1 + z_log_var - tf.square(z_mean) - tf.exp(z_log_var)
        kl_loss = tf.reduce_mean(kl_loss)
        kl_loss *= -0.5
        self.add_loss(reconstruction_loss + kl_loss)
        return reconstructed

# Encoder
latent_dim = 2

encoder_inputs = keras.Input(shape=(200, 200, 3))
x = layers.Conv2D(32, 3, activation="relu", strides=2, padding="same")(encoder_inputs)
x = layers.Conv2D(64, 3, activation="relu", strides=2, padding="same")(x)
x = layers.Flatten()(x)
x = layers.Dense(16, activation="relu")(x)
z_mean = layers.Dense(latent_dim, name="z_mean")(x)
z_log_var = layers.Dense(latent_dim, name="z_log_var")(x)
z = layers.Lambda(lambda args: sampling(*args), output_shape=(latent_dim,), name='z')([z_mean, z_log_var])

encoder = keras.Model(encoder_inputs, [z_mean, z_log_var, z], name='encoder')
encoder.summary()

Model: "encoder"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ Layer (type)                  ┃ Output Shape              ┃         Param # ┃ Connected to               ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩
│ input_layer (InputLayer)      │ (None, 200, 200, 3)       │               0 │ -                          │
├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤
│ conv2d (Conv2D)               │ (None, 100, 100, 32)      │             896 │ input_layer[0][0]          │
├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤
│ conv2d_1 (Conv2D)             │ (None, 50, 50, 64)        │          18,496 │ conv2d[0][0]               │
├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤
│ flatten (Flatten)             │ (None, 160000)            │               0 │ conv2d_1[0][0]             │
├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤
│ dense (Dense)                 │ (None, 16)                │       2,560,016 │ flatten[0][0]              │
├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤
│ z_mean (Dense)                │ (None, 2)                 │              34 │ dense[0][0]                │
├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤
│ z_log_var (Dense)             │ (None, 2)                 │              34 │ dense[0][0]                │
├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤
│ z (Lambda)                    │ (None, 2)                 │               0 │ z_mean[0][0],              │
│                               │                           │                 │ z_log_var[0][0]            │
└───────────────────────────────┴───────────────────────────┴─────────────────┴────────────────────────────┘

 Total params: 2,579,476 (9.84 MB)

 Trainable params: 2,579,476 (9.84 MB)

 Non-trainable params: 0 (0.00 B)

# Decoder
latent_inputs = keras.Input(shape=(latent_dim,))
x = layers.Dense(50 * 50 * 64, activation="relu")(latent_inputs)
x = layers.Reshape((50, 50, 64))(x)
x = layers.Conv2DTranspose(64, 3, activation="relu", strides=2, padding="same")(x)
x = layers.Conv2DTranspose(32, 3, activation="relu", strides=2, padding="same")(x)
decoder_outputs = layers.Conv2DTranspose(3, 3, activation="sigmoid", padding="same")(x)

decoder = keras.Model(latent_inputs, decoder_outputs, name='decoder')
decoder.summary()

Model: "decoder"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓
┃ Layer (type)                         ┃ Output Shape                ┃         Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩
│ input_layer_1 (InputLayer)           │ (None, 2)                   │               0 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ dense_1 (Dense)                      │ (None, 160000)              │         480,000 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ reshape (Reshape)                    │ (None, 50, 50, 64)          │               0 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ conv2d_transpose (Conv2DTranspose)   │ (None, 100, 100, 64)        │          36,928 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ conv2d_transpose_1 (Conv2DTranspose) │ (None, 200, 200, 32)        │          18,464 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ conv2d_transpose_2 (Conv2DTranspose) │ (None, 200, 200, 3)         │             867 │
└──────────────────────────────────────┴─────────────────────────────┴─────────────────┘

 Total params: 536,259 (2.05 MB)

 Trainable params: 536,259 (2.05 MB)

 Non-trainable params: 0 (0.00 B)

# VAE Model
vae = VAE(encoder, decoder)
vae.compile(optimizer='adam')
vae.build(input_shape=(None, 200, 200, 3))
vae.summary()

# Prepare and normalize the image
pic_1 = keras.preprocessing.image.load_img('pic_1.jpeg', target_size=(200, 200))
pic_1 = keras.preprocessing.image.img_to_array(pic_1).astype("float32") / 255
pic_1 = np.expand_dims(pic_1, 0)

# Train the VAE model
vae.fit(pic_1, epochs=100, batch_size=1)

Model: "vae"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓
┃ Layer (type)                         ┃ Output Shape                ┃         Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩
│ encoder (Functional)                 │ ?                           │       2,579,476 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ decoder (Functional)                 │ ?                           │         536,259 │
└──────────────────────────────────────┴─────────────────────────────┴─────────────────┘

 Total params: 3,115,735 (11.89 MB)

 Trainable params: 3,115,735 (11.89 MB)

 Non-trainable params: 0 (0.00 B)

Epoch 1/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 2s 2s/step - loss: 83177.3750
Epoch 2/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 56ms/step - loss: 83199.9844
Epoch 3/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 57ms/step - loss: 83137.7109
Epoch 4/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 60ms/step - loss: 83081.3359
Epoch 5/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 58ms/step - loss: 83006.4609
Epoch 6/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 58ms/step - loss: 82808.2734
Epoch 7/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 61ms/step - loss: 82488.0391
Epoch 8/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 59ms/step - loss: 82183.8203
Epoch 9/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 55ms/step - loss: 81755.2578
Epoch 10/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 56ms/step - loss: 80598.4375
Epoch 11/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 55ms/step - loss: 79338.1484
Epoch 12/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 58ms/step - loss: 78044.2891
Epoch 13/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 59ms/step - loss: 76914.7656
Epoch 14/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 61ms/step - loss: 75710.9375
Epoch 15/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 58ms/step - loss: 74347.7656
Epoch 16/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 56ms/step - loss: 73219.6562
Epoch 17/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 54ms/step - loss: 72456.3281
Epoch 18/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 58ms/step - loss: 71910.6094
Epoch 19/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 56ms/step - loss: 71444.3828
Epoch 20/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 57ms/step - loss: 71049.1016
Epoch 21/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 56ms/step - loss: 70710.1562
Epoch 22/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 56ms/step - loss: 70397.3125
Epoch 23/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 55ms/step - loss: 70146.9375
Epoch 24/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 56ms/step - loss: 69991.2422
Epoch 25/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 78ms/step - loss: 69869.0469
Epoch 26/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 74ms/step - loss: 69733.7891
Epoch 27/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 59ms/step - loss: 69622.5547
Epoch 28/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 59ms/step - loss: 69523.6719
Epoch 29/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 59ms/step - loss: 69410.1484
Epoch 30/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 59ms/step - loss: 69317.9375
Epoch 31/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 88ms/step - loss: 69252.4141
Epoch 32/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 67ms/step - loss: 69166.5156
Epoch 33/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 61ms/step - loss: 69107.1328
Epoch 34/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 57ms/step - loss: 69046.4141
Epoch 35/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 59ms/step - loss: 68993.1875
Epoch 36/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 55ms/step - loss: 68969.3984
Epoch 37/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 55ms/step - loss: 68947.7969
Epoch 38/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 63ms/step - loss: 68869.1641
Epoch 39/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 60ms/step - loss: 68868.7969
Epoch 40/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 59ms/step - loss: 68825.5625
Epoch 41/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 58ms/step - loss: 68771.5547
Epoch 42/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 58ms/step - loss: 68728.8594
Epoch 43/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 56ms/step - loss: 68762.8047
Epoch 44/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 56ms/step - loss: 68681.9766
Epoch 45/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 58ms/step - loss: 68623.1094
Epoch 46/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 53ms/step - loss: 68669.6094
Epoch 47/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 60ms/step - loss: 68528.6094
Epoch 48/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 57ms/step - loss: 68514.2266
Epoch 49/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 58ms/step - loss: 68489.3438
Epoch 50/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 59ms/step - loss: 68406.0469
Epoch 51/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 59ms/step - loss: 68353.9297
Epoch 52/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 66ms/step - loss: 68332.2188
Epoch 53/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 59ms/step - loss: 68261.9141
Epoch 54/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 59ms/step - loss: 68183.1953
Epoch 55/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 61ms/step - loss: 68140.0703
Epoch 56/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 60ms/step - loss: 68095.0469
Epoch 57/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 61ms/step - loss: 67997.8047
Epoch 58/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 61ms/step - loss: 67951.8047
Epoch 59/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 58ms/step - loss: 67896.7266
Epoch 60/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 56ms/step - loss: 67806.6484
Epoch 61/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 59ms/step - loss: 67762.4375
Epoch 62/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 56ms/step - loss: 67698.6406
Epoch 63/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 55ms/step - loss: 67666.8828
Epoch 64/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 54ms/step - loss: 67597.9141
Epoch 65/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 59ms/step - loss: 67606.9609
Epoch 66/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 60ms/step - loss: 67540.2578
Epoch 67/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 57ms/step - loss: 67496.0938
Epoch 68/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 56ms/step - loss: 67560.3906
Epoch 69/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 59ms/step - loss: 67456.5078
Epoch 70/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 58ms/step - loss: 67455.4766
Epoch 71/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 58ms/step - loss: 67439.9688
Epoch 72/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 56ms/step - loss: 67391.3828
Epoch 73/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 56ms/step - loss: 67367.6094
Epoch 74/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 60ms/step - loss: 67405.2031
Epoch 75/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 59ms/step - loss: 67329.3984
Epoch 76/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 57ms/step - loss: 67312.9844
Epoch 77/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 60ms/step - loss: 67310.9297
Epoch 78/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 101ms/step - loss: 67284.5234
Epoch 79/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 58ms/step - loss: 67261.4219
Epoch 80/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 57ms/step - loss: 67256.9219
Epoch 81/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 56ms/step - loss: 67239.6406
Epoch 82/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 55ms/step - loss: 67215.1641
Epoch 83/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 55ms/step - loss: 67207.1094
Epoch 84/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 55ms/step - loss: 67196.3984
Epoch 85/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 53ms/step - loss: 67181.4922
Epoch 86/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 57ms/step - loss: 67167.9766
Epoch 87/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 55ms/step - loss: 67180.4688
Epoch 88/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 58ms/step - loss: 67152.2734
Epoch 89/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 58ms/step - loss: 67145.4922
Epoch 90/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 61ms/step - loss: 67135.4688
Epoch 91/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 59ms/step - loss: 67125.2109
Epoch 92/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 57ms/step - loss: 67115.7500
Epoch 93/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 59ms/step - loss: 67105.8438
Epoch 94/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 56ms/step - loss: 67106.8203
Epoch 95/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 54ms/step - loss: 67091.8750
Epoch 96/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 59ms/step - loss: 67082.9375
Epoch 97/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 56ms/step - loss: 67074.0703
Epoch 98/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 60ms/step - loss: 67084.1484
Epoch 99/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 60ms/step - loss: 67063.2578
Epoch 100/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 57ms/step - loss: 67076.4297

<keras.src.callbacks.history.History at 0x1c974460b50>

# Show plot and save image function
def show_and_save_plot(image, save_path):
    plt.imshow(image.squeeze())
    plt.axis('off')
    plt.savefig(save_path)
    plt.show()


for i in range(10):
    z_mean, z_log_var, _ = vae.encoder.predict(pic_1)
    # Introduce randomness in sampling
    encoded_imgs = sampling(z_mean, z_log_var).numpy()
    decoded_imgs = vae.decoder.predict(encoded_imgs)
    save_path = f'generated_image_{i + 1}.png'
    show_and_save_plot(decoded_imgs, save_path)

print("Images have been saved and displayed successfully.")

1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 93ms/step
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 72ms/step

1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 23ms/step
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step

1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 23ms/step
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 34ms/step

1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 27ms/step

1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 22ms/step
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step

1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 34ms/step
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 18ms/step

1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 23ms/step
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 22ms/step

1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 22ms/step
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 22ms/step

1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 19ms/step
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 18ms/step

1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 21ms/step

Images have been saved and displayed successfully.

What output do you see? Explain why

For the codes above, why when it is generating black/white 10 images, the 10 images colors can be different but when it is generating colorful 10 images, the 10 images look the same?

z_mean, z_log_var, _ = vae.encoder.predict(pic_1)
print("z_mean:", z_mean)
print("z_log_var:", z_log_var)

1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 24ms/step
z_mean: [[-12.294494   -5.9670925]]
z_log_var: [[-10.742041   -2.4054945]]

Based on the what and why analysis above, I adjusted my parameters below.

Now the newly generated 10 images are much different!

import numpy as np
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers
import matplotlib.pyplot as plt

# Sampling function with randomness
def sampling(z_mean, z_log_var):
    batch = tf.shape(z_mean)[0]
    dim = tf.shape(z_mean)[1]
    epsilon = tf.keras.backend.random_normal(shape=(batch, dim))
    return z_mean + tf.exp(0.5 * z_log_var) * epsilon

# VAE Class with custom call method
class VAE(keras.Model):
    def __init__(self, encoder, decoder, **kwargs):
        super(VAE, self).__init__(**kwargs)
        self.encoder = encoder
        self.decoder = decoder
        self.mse = tf.keras.losses.MeanSquaredError()

    def call(self, inputs):
        z_mean, z_log_var, z = self.encoder(inputs)
        reconstructed = self.decoder(z)
        reconstruction_loss = tf.reduce_mean(
            self.mse(inputs, reconstructed)
        )
        reconstruction_loss *= 200 * 200 * 3
        kl_loss = 1 + z_log_var - tf.square(z_mean) - tf.exp(z_log_var)
        kl_loss = tf.reduce_mean(kl_loss)
        kl_loss *= -0.5
        self.add_loss(reconstruction_loss + kl_loss)
        return reconstructed

# Encoder
latent_dim = 10  # Increased latent dimension

encoder_inputs = keras.Input(shape=(200, 200, 3))
x = layers.Conv2D(32, 3, activation="relu", strides=2, padding="same")(encoder_inputs)
x = layers.Conv2D(64, 3, activation="relu", strides=2, padding="same")(x)
x = layers.Flatten()(x)
x = layers.Dense(16, activation="relu")(x)
z_mean = layers.Dense(latent_dim, name="z_mean")(x)
z_log_var = layers.Dense(latent_dim, name="z_log_var")(x)
z = layers.Lambda(lambda args: sampling(*args), output_shape=(latent_dim,), name='z')([z_mean, z_log_var])

encoder = keras.Model(encoder_inputs, [z_mean, z_log_var, z], name='encoder')
encoder.summary()

# Decoder
latent_inputs = keras.Input(shape=(latent_dim,))
x = layers.Dense(50 * 50 * 64, activation="relu")(latent_inputs)
x = layers.Reshape((50, 50, 64))(x)
x = layers.Conv2DTranspose(64, 3, activation="relu", strides=2, padding="same")(x)
x = layers.Conv2DTranspose(32, 3, activation="relu", strides=2, padding="same")(x)
decoder_outputs = layers.Conv2DTranspose(3, 3, activation="sigmoid", padding="same")(x)

decoder = keras.Model(latent_inputs, decoder_outputs, name='decoder')
decoder.summary()

# VAE Model
vae = VAE(encoder, decoder)
vae.compile(optimizer='adam')
vae.build(input_shape=(None, 200, 200, 3))
vae.summary()

# Prepare and normalize the image
pic_1 = keras.preprocessing.image.load_img('pic_1.jpeg', target_size=(200, 200))
pic_1 = keras.preprocessing.image.img_to_array(pic_1).astype("float32") / 255
pic_1 = np.expand_dims(pic_1, 0)

# Train the VAE model
vae.fit(pic_1, epochs=100, batch_size=1)

# Show plot and save image function
def show_and_save_plot(image, save_path):
    plt.imshow(image.squeeze())
    plt.axis('off')
    plt.savefig(save_path)
    plt.show()

# Run the encoder-decoder multiple times to generate different copies
for i in range(10):
    z_mean, z_log_var, _ = vae.encoder.predict(pic_1)
    # Introduce randomness in sampling
    encoded_imgs = sampling(z_mean, z_log_var).numpy()
    decoded_imgs = vae.decoder.predict(encoded_imgs)
    save_path = f'generated_image_{i + 1}.png'
    show_and_save_plot(decoded_imgs, save_path)

print("Images have been saved and displayed successfully.")

Model: "encoder"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ Layer (type)                  ┃ Output Shape              ┃         Param # ┃ Connected to               ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩
│ input_layer_2 (InputLayer)    │ (None, 200, 200, 3)       │               0 │ -                          │
├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤
│ conv2d_2 (Conv2D)             │ (None, 100, 100, 32)      │             896 │ input_layer_2[0][0]        │
├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤
│ conv2d_3 (Conv2D)             │ (None, 50, 50, 64)        │          18,496 │ conv2d_2[0][0]             │
├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤
│ flatten_1 (Flatten)           │ (None, 160000)            │               0 │ conv2d_3[0][0]             │
├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤
│ dense_2 (Dense)               │ (None, 16)                │       2,560,016 │ flatten_1[0][0]            │
├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤
│ z_mean (Dense)                │ (None, 10)                │             170 │ dense_2[0][0]              │
├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤
│ z_log_var (Dense)             │ (None, 10)                │             170 │ dense_2[0][0]              │
├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤
│ z (Lambda)                    │ (None, 10)                │               0 │ z_mean[0][0],              │
│                               │                           │                 │ z_log_var[0][0]            │
└───────────────────────────────┴───────────────────────────┴─────────────────┴────────────────────────────┘

 Total params: 2,579,748 (9.84 MB)

 Trainable params: 2,579,748 (9.84 MB)

 Non-trainable params: 0 (0.00 B)

Model: "decoder"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓
┃ Layer (type)                         ┃ Output Shape                ┃         Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩
│ input_layer_3 (InputLayer)           │ (None, 10)                  │               0 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ dense_3 (Dense)                      │ (None, 160000)              │       1,760,000 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ reshape_1 (Reshape)                  │ (None, 50, 50, 64)          │               0 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ conv2d_transpose_3 (Conv2DTranspose) │ (None, 100, 100, 64)        │          36,928 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ conv2d_transpose_4 (Conv2DTranspose) │ (None, 200, 200, 32)        │          18,464 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ conv2d_transpose_5 (Conv2DTranspose) │ (None, 200, 200, 3)         │             867 │
└──────────────────────────────────────┴─────────────────────────────┴─────────────────┘

 Total params: 1,816,259 (6.93 MB)

 Trainable params: 1,816,259 (6.93 MB)

 Non-trainable params: 0 (0.00 B)

Model: "vae_1"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓
┃ Layer (type)                         ┃ Output Shape                ┃         Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩
│ encoder (Functional)                 │ ?                           │       2,579,748 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ decoder (Functional)                 │ ?                           │       1,816,259 │
└──────────────────────────────────────┴─────────────────────────────┴─────────────────┘

 Total params: 4,396,007 (16.77 MB)

 Trainable params: 4,396,007 (16.77 MB)

 Non-trainable params: 0 (0.00 B)

Epoch 1/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 2s 2s/step - loss: 7610.7393
Epoch 2/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 67ms/step - loss: 7726.1265
Epoch 3/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 70ms/step - loss: 7600.4229
Epoch 4/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 69ms/step - loss: 7594.1895
Epoch 5/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 70ms/step - loss: 7586.2324
Epoch 6/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 75ms/step - loss: 7585.9980
Epoch 7/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 74ms/step - loss: 7580.5942
Epoch 8/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 72ms/step - loss: 7562.7554
Epoch 9/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 70ms/step - loss: 7562.2412
Epoch 10/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 66ms/step - loss: 7558.2412
Epoch 11/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 69ms/step - loss: 7538.1782
Epoch 12/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 83ms/step - loss: 7530.4229
Epoch 13/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 77ms/step - loss: 7481.6230
Epoch 14/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 108ms/step - loss: 7464.0249
Epoch 15/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 92ms/step - loss: 7488.7236
Epoch 16/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 79ms/step - loss: 7442.9277
Epoch 17/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 69ms/step - loss: 7459.9717
Epoch 18/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 97ms/step - loss: 7406.3745
Epoch 19/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 172ms/step - loss: 7363.2588
Epoch 20/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 71ms/step - loss: 7301.2778
Epoch 21/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 80ms/step - loss: 7249.5308
Epoch 22/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 75ms/step - loss: 7515.9595
Epoch 23/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 92ms/step - loss: 7032.7549
Epoch 24/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 83ms/step - loss: 7185.9648
Epoch 25/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 78ms/step - loss: 6800.1313
Epoch 26/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 71ms/step - loss: 7498.6348
Epoch 27/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 87ms/step - loss: 7238.9546
Epoch 28/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 78ms/step - loss: 6345.5332
Epoch 29/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 77ms/step - loss: 6500.5693
Epoch 30/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 68ms/step - loss: 6510.3496
Epoch 31/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 84ms/step - loss: 6107.6074
Epoch 32/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 90ms/step - loss: 5453.5703
Epoch 33/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 87ms/step - loss: 5963.7236
Epoch 34/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 86ms/step - loss: 5101.7266
Epoch 35/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 92ms/step - loss: 5312.6353
Epoch 36/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 81ms/step - loss: 5710.2451
Epoch 37/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 69ms/step - loss: 4759.2271
Epoch 38/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 62ms/step - loss: 4363.7920
Epoch 39/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 70ms/step - loss: 5299.1719
Epoch 40/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 66ms/step - loss: 4659.5791
Epoch 41/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 67ms/step - loss: 4278.9546
Epoch 42/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 69ms/step - loss: 4238.8711
Epoch 43/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 62ms/step - loss: 3939.6272
Epoch 44/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 66ms/step - loss: 4250.3989
Epoch 45/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 65ms/step - loss: 3802.0754
Epoch 46/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 66ms/step - loss: 4096.0186
Epoch 47/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 66ms/step - loss: 3790.0337
Epoch 48/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 84ms/step - loss: 4175.6968
Epoch 49/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 99ms/step - loss: 4180.9517
Epoch 50/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 67ms/step - loss: 4046.8005
Epoch 51/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 70ms/step - loss: 3911.4573
Epoch 52/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 67ms/step - loss: 3511.0881
Epoch 53/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 65ms/step - loss: 3465.3376
Epoch 54/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 63ms/step - loss: 3440.6035
Epoch 55/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 66ms/step - loss: 3532.6001
Epoch 56/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 62ms/step - loss: 3305.6157
Epoch 57/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 66ms/step - loss: 3245.9075
Epoch 58/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 62ms/step - loss: 3229.1716
Epoch 59/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 64ms/step - loss: 3110.1326
Epoch 60/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 64ms/step - loss: 3692.0945
Epoch 61/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 65ms/step - loss: 2908.8545
Epoch 62/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 65ms/step - loss: 3116.0874
Epoch 63/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 64ms/step - loss: 2801.7007
Epoch 64/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 66ms/step - loss: 2918.8108
Epoch 65/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 65ms/step - loss: 2768.5195
Epoch 66/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 67ms/step - loss: 2687.1475
Epoch 67/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 63ms/step - loss: 2752.2695
Epoch 68/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 69ms/step - loss: 3081.9973
Epoch 69/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 64ms/step - loss: 2474.6143
Epoch 70/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 64ms/step - loss: 2219.4907
Epoch 71/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 66ms/step - loss: 2633.3442
Epoch 72/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 65ms/step - loss: 2011.7100
Epoch 73/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 69ms/step - loss: 1995.2678
Epoch 74/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 64ms/step - loss: 2160.3147
Epoch 75/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 66ms/step - loss: 2189.7275
Epoch 76/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 67ms/step - loss: 1904.3241
Epoch 77/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 74ms/step - loss: 1769.3903
Epoch 78/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 75ms/step - loss: 1916.2294
Epoch 79/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 71ms/step - loss: 2144.8218
Epoch 80/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 65ms/step - loss: 2341.8989
Epoch 81/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 66ms/step - loss: 1934.7032
Epoch 82/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 66ms/step - loss: 1598.4662
Epoch 83/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 70ms/step - loss: 2005.9113
Epoch 84/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 73ms/step - loss: 2034.1517
Epoch 85/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 75ms/step - loss: 3048.6072
Epoch 86/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 72ms/step - loss: 2552.8713
Epoch 87/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 63ms/step - loss: 1493.8463
Epoch 88/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 65ms/step - loss: 1886.9950
Epoch 89/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 65ms/step - loss: 1611.8395
Epoch 90/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 66ms/step - loss: 2299.6582
Epoch 91/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 77ms/step - loss: 1591.7374
Epoch 92/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 74ms/step - loss: 1661.3323
Epoch 93/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 71ms/step - loss: 2869.6941
Epoch 94/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 65ms/step - loss: 1418.2002
Epoch 95/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 62ms/step - loss: 1554.6145
Epoch 96/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 66ms/step - loss: 1910.1365
Epoch 97/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 64ms/step - loss: 1320.2623
Epoch 98/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 65ms/step - loss: 1249.3489
Epoch 99/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 76ms/step - loss: 1830.5522
Epoch 100/100
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 92ms/step - loss: 1435.1040
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 91ms/step
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 85ms/step

1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 24ms/step
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 21ms/step

1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 27ms/step
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 21ms/step

1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 17ms/step
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step

1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 36ms/step

1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 24ms/step

1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step

1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 19ms/step
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step

1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 35ms/step
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 19ms/step

1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 19ms/step

Images have been saved and displayed successfully.